US 20050132129 A1
A method is disclosed for emulating a disk drive on flash memory, thus enabling one or more file-systems to be resident on flash memory. A list of data structures is maintained corresponding to certain sectors resident on flash memory. Certain characteristics of data access to those sectors are recorded and a weightage value is dynamically computed, which is used to select those sectors that remain cached. This weightage value is calculated so that it minimizes erases and speeds up writes to flash memory and also automatically detects flash memory sectors that hold critical meta-data information or frequently accessed files and gives preferential treatment to those sectors. The number of sector caches available to the device driver is configured depending on the memory available in the system.
1. A method of managing memory contents of a flash memory having a plurality of memory sectors that are erased before being rewritten, the method comprising the steps of caching selected memory sectors that hold file-system blocks, computing a sector weightage for each cached memory sector that is used to determine which memory sectors remain cached, wherein the sector weightage is dependent upon a record of characteristics of data accesses to the memory sector.
2. The method as claimed in
3. The method as claimed in
4. The method as claimed in
5. The method as claimed in
6. The method as claimed in
7. The method as claimed in
8. The method as claimed in
9. The method as claimed in
10. A method of managing data in flash memory having a plurality of memory sectors that are erased before being rewritten, comprising the steps of:
caching selected memory sectors;
receiving a write request to perform a write operation in a memory sector;
computing a sector weightage for each cached memory sector that is used to determine which memory sectors remain cached, wherein the sector weightage is dependent upon a record of characteristics of data accesses to the memory sector;
writing the data to a sector cache corresponding to the memory sector of said received write request, if the stored memory sector subject of said received write request is already stored in the sector cache; and
writing the data to an available free sector cache if the memory sector of said received write request is not already stored in a sector cache.
11. The method as claimed in
selecting an unpinned stored memory sector on the basis of the sector cache that currently has the lowest calculated sector weightage of all the sector caches;
setting the sector weightage and number of sector switches for the selected sector cache to be equal to that of a sector cache that currently has the minimum sector weightage among all other sector caches;
erasing the selected sector from the memory and writing the contents of the corresponding sector cache to said selected memory sector; and
reading the contents of the memory sector for which the write request was received, into the selected sector cache and writing the requested data onto the selected sector cache.
12. The method as claimed in
13. The method as claimed in
The present invention relates to the use of flash memory as a data storage device and relates particularly though not exclusively to improved techniques of data management using flash memory, especially when the flash memory is accessed from the operating system of an embedded device for which the flash memory acts as a data storage device.
The contents of flash memory are updated following a preceding sector erase of the relevant area. Conventional storage mediums do not generally require such erasures. Accordingly, there is a findamental incompatibility in the use of flash memory as a data storage medium with computer systems using conventional operating systems.
One method for storing data on flash memory is disclosed in U.S. Pat. No. 5,404,485 entitled “Flash File System”, issued to Ban. In this case, a virtual mapping system is provided that allows data to be continuously written to unwritten physical address locations. The method used by Suzuki (US 2002/0166022 A1) also uses logical to physical address mappings. When virtual or logical memory maps are used, the use of standard file-system utilities to debug/repair the stored file-system is precluded when the file-system data is uploaded from an embedded device that uses flash memory and is moved to a customer support site. The method described as part of Ogawa (U.S. Pat. No. 5,933,847) uses the Least Recently Used (LRU) algorithm to select cached sectors that can be freed. This will not be optimal in many cases when flash memory emulates a disk drive in an embedded device. Also, these techniques do not automatically detect critical meta-data sectors and offer them persistent storage or preferential treatment.
In view of the above, a need exists for a manner of managing data in flash memory that addresses certain limitations of prior art.
Embedded devices typically have flash memory banks for permanent storage. The present invention provides a method for emulating a disk drive on flash memory, thus enabling one or more file-systems to be resident on flash memory. A list of data structures is maintained corresponding to certain sectors resident on flash memory. Certain characteristics of data access to those sectors are recorded and a weightage value is dynamically computed, which is used to select those sectors that can remain cached. This weightage value is algorithmically calculated so that it will minimize erases and speed up writes to flash and also automatically detect flash sectors that hold critical meta-data information or frequently accessed files and gives preferential treatment to those sectors.
No allocation data structures need to be stored on flash memory. One can upload the embedded flash file-system image to an external host machine and use standard file-system utilities to debug/repair the file-system on the host machine.
The described techniques are implemented as a device driver for a flash memory device. Data is written onto the sector caches rather than the flash-disk, and depending on the dynamically calculated sector weightage, swapped back and forth between the sector caches and flash memory sectors, on demand.
A method, device driver and computer system are described for providing an advantageous scheme for accessing (reading from and writing to) the contents of flash memory from a computer operating system. A device driver essentially converts the more general input/output instructions of the operating system installed on the computer to messages that the device can interpret, to allow the device and the operating system to communicate with each other. In this case, a flash memory is a device intended to be attached to a computer, so that a described device driver can be installed in the operating system to allow the flash memory to act as a data storage medium for the computer.
Flash memory is normally organized into banks and further into sectors. Erases of selected portions of the flash memory are performed at the granularity of individual sectors. A flash-write to a location is preceded by a corresponding flash-erase.
The flash-disk device driver maps all the flash banks into memory that is virtually contiguous, i.e., the banks are mapped adjacent to one another in virtual memory.
Since the described algorithm is implemented at the device driver level, it is transparent to the file-system layer of the operating system. The algorithm assumes that the embedded operating system uses file-system buffer caching algorithms, such as those generally used by UNIX operating systems. The algorithm is based on caching flash sectors depending on a dynamically calculated weightage value based on the characteristics of data access to those sectors.
A brief description of terminology used herein is given below.
Flash memory: A type of non-volatile memory which is an electrically erasable and programmable read-only memory (EEPROM) having a programmable operation, which allows for the erasure of blocks of memory.
Flash-disk: The area in flash memory that is available for use to emulate a disk.
Flash-disk driver: The block device driver for the flash-disk. The algorithm described in this discussion is implemented by the flash-disk block driver.
Flash-disk sector: A physical sector in the flash-disk.
Sector cache: A cached copy of a flash-disk sector, maintained by the flash-disk driver. A sector cache can be pinned or unpinned.
Sector cache list: A linked list consisting of the sector caches maintained by the flash-disk driver.
Sector switch: A flash write requires a sector switch if the previous write was not done to the same flash sector.
Minor devices: Instances of the flash-disk driver—there is one instance per file-system resident on the flash-disk.
Buffer-cache: A layer of software present in UNIX systems, that reside between the file-system and the device driver. Read and write requests from the buffer-cache to the disk are normally issued in terms of file-system blocks of data.
Sync: An operation that flushes the dirtied blocks in the buffer-cache to the disk. The “sync” (for synchronization) operation might be done at periodic intervals, but definitely during system shutdown. Writes to the disk may also occur when the buffer cache gets shrunk when the system runs low on memory.
Super-block: The portion of the file-system that contains meta-data information.
The flash memory device driver maintains a linked list of sector caches. The number of sector caches available to the flash memory driver is configured depending on the memory available in the system. Data is written onto the sector caches, rather than the flash-disk device, and depending on a dynamically calculated weightage value, swapped back and forth between the sector caches and flash memory sectors, on demand.
When the flash-disk driver needs to write to a flash sector, it first sees if the sector is already available in the sector cache list. If the sector is available, the write is done onto the corresponding sector cache. If the sector is not available, and if there is a free sector available in the cache list, that free sector is populated with the corresponding flash sector contents and the write is done onto this sector cache. If the sector is not available and there are no free sector caches in the list, a sector cache has to be selected to be freed, by swapping the contents of the selected sector cache back to the flash-disk.
A sector weightage is dynamically assigned to each sector that is cached. This count is incremented whenever there is a write access to a file-system block within this sector cache, if the block has not already caused the sector weightage to be incremented. This can be tracked by using one bit per file-system block present in the sector. All tracking bits are initially reset. Whenever a block is written to, the corresponding bit is set. As will be explained later, these tracking bits will be reset and the sector weightage set to one greater than the maximum sector weightage of all cached sectors, when the algorithm determines that this sector has suffered sector switches greater than a pre-calculated threshold value.
A sector cache that is marked as “pinned”, cannot be swapped back to disk and reused (as described below in the section entitled “Pinning a sector cache”). Among the other sector caches, the cache that has the minimum sector weightage is selected to be swapped out to the flash-disk. A flash-disk will typically be used as a storage medium in small embedded devices. While the program binaries stored in these devices (which are read-only) could be large in size, the files that the user wishes to modify will be usually relatively small in size. So a sector could be holding multiple write-able files. If more blocks in a sector cache have been modified, it generally means that more files resident on that sector have been modified—and at least one of them might be accessed again in the near future.
When a sector cache is to be swapped-out to flash-disk, the corresponding sector in the flash-disk is erased and rewritten with the sector cache's contents. The data in the flash-disk sector for which the write request has been received, is then swapped-in to this sector cache. Since this sector cache now contains the most recently used file-system block in the flash-disk, we reward it with an advantage by setting its sector weightage and number of sector switches equal to that of the sector cache that has the minimum sector weightage among all other sector caches.
Automatically Detecting Sectors Holding Meta-Data
Whenever a file gets written to disk from the Operating System buffer cache (usually during a sync), meta-data information for the file is updated in the file-system super-block. The sector(s) containing the super block are usually the most frequently accessed sectors in the flash-disk. The meta-data sectors thus have more number of sector switches (a sector cache write is said to require a sector switch, if the previous write was not done to the same sector cache), and the number of modified blocks during each such switch is less (since the space required for meta-data for a file is usually less).
Whenever the number of sector switches to a sector cache reaches a threshold, the sector weightage for that sector cache is set to one greater than the maximum sector weightage of all the cached sectors. When the threshold is reached, the number of sector switches for the sector cache is also reset to zero. This means that, this sector cache will enjoy an advantage compared to other sector caches for subsequent swap decisions, since a sector cache with the least sector weightage is swapped-out first.
If the sector weightage of a sector is greater than all other sectors, the sector cache can remain in the sector cache list, until the sector weightage, becomes less than the sector weightage of all other sector caches. Thus, when a sector cache demonstrates the characteristics of a sector cache that holds meta-data information (that is, by having a large number of switches to the sector cache) the sector cache is rewarded with this advantage.
A sector that caches file-system data may have characteristics similar to a sector caching file-system meta-data, if the sector that caches file-system data contains a large number of very small files, many of which have been modified. Such sectors also get rewarded with the bonus sector weightage as long as they continuously get accessed. They will eventually get swapped-out to the flash-disk since only a sector that has continuous and frequent write accesses persists in the sector cache list.
Calculating the Sector Switch Threshold
The value of the sector switch threshold can be decided, taking into account, the size of a sector cache, the average size of write-able files in the embedded device, and other factors relevant to the device. So, an approximation for the threshold can be given by the mathematical expression (T=K×m/n) which expresses the mathematic product of K and (m/n), in which T is the threshold for the number of switches to a sector, m is the size of the flash sectors, n is the average size of write-able files in the device and K is a constant numerical value. For smaller values of K, the calculated threshold is more conservative (that is, also smaller).
Pinning a Sector Cache
To guarantee the permanent presence of certain critical sectors in cache, those sectors can be pinned in memory. A pinned sector cache is not available for further swapping. It is permanently dedicated to the flash-disk sector that it is caching.
The file-system meta-data region normally resides in the initial portions of the file-system image. The size of this region depends on the particular file system that is used, the file-system size, the size of the flash sectors holding the initial portions of the file-system and the amount of memory used by the file-system in the first sector, if the file-system does not start on a sector boundary.
Multiple file-systems can be accommodated on the flash-disk. The flash-disk driver is able to support multiple devices—one per file-system. That is, multiple flash-disks can be used in conjunction with a suitable computer system. In the terminology of UNIX operating systems, the multiple flash-disks are minor devices. In this case, the sector cache list data structure is global to all minor devices.
If multiple file-system fragments are present (as is the case in
Procedure for Flash Writes
When the flash-disk device driver receives a request to write a block of data (a write request), the driver performs a sequence of steps. These steps, and the algorithm governing these steps is described below with reference to
When the flash-disk driver receives a read request for a block of data, it does the following:
The above sequence is as one would expect, and simply operates to access the most recent information recorded in memory, either in the flash memory or the associated cache.
A sector map array is also used, in which each entry (one per sector per bank) points to a sector cache in the above list if it is currently cached, or to
For a write operation to a flash sector, direct indexing is done onto the sector map array, to determine whether it is currently present in the sector cache list. If it is (that is, the corresponding entry in the map array is non-NULL), the corresponding sector cache is used for the flash write. Otherwise the sector cache information (sc_info) list is traversed in order to determine whether a free cache entry is available in the list (this is done by checking if a pre-determined invalid value is found in the bank/sector number field of any sc_info element in the list). Also during this traversal, the address of the sc_info element with the lowest sector weightage is found and stored, provided that a free sc_info element has not been found. If the sector is not in the cache list and a free sc_info element is available in the list (as determined in the last step), the free entry is populated with the corresponding flash sector's contents and the flash-write is performed onto that sector. If there is no free cache element, the sc_info element with the lowest sector weightage (as determined in the last step) is selected to be swapped back to the flash-disk and thus the freed entry is populated with the corresponding flash-sector and the write is performed onto this sector. The corresponding sc_info element is updated with new values. The corresponding pointers in the map array for the old and new sector are also updated.
The size of the sector caches in memory is equal to the size of the largest sector in the flash-disk. The memory for the sector caches is allocated during initialization of the flash-disk driver. The number of sector caches in the list can be configured at compile-time of the flash-disk driver, keeping in mind the memory available to the operating system. If the flash memory has variable sized sectors, a sector cache can be shared by more than one smaller sector.
This algorithm operates satisfactorily under the assumption that the system is shutdown as intended. Under this assumption, further improvements can be effected, like kicking off a thread for the flash-disk sector erase, just after the sector cache is swapped-in (instead of erasing just before a swap-out). If the assumption of regular shutdowns does not hold, periodically flushing dirtied sector caches to the flash-disk adds some robustness.
Computer Hardware and Software
The above-described technique can be implemented using a device driver 620 in conjunction with an operating system 630 installed and executing on a host computer system 700.
The computer 750 includes the control module 766, a memory 770 that may include random access memory (RAM) and read-only memory (ROM), input/output (I/O) interfaces 764, 772, a video interface 760, and one or more storage devices generally represented by the storage device 762. The control module 766 is implemented using a central processing unit (CPU) that executes or runs a computer readable software program code that performs a particular function or related set of functions.
Each of the elements in the computer system 750 is typically connected to other devices via a bus 780 that in turn can consist of data, address, and control buses. The video interface 760 is connected to the video display 710 and provides video signals from the computer 750 for display on the video display 710. User input to operate the computer 750 could be provided by one or more of the input devices 730, 732 via the I/O interface 772. For example, a user of the computer 750 can use a keyboard as I/O interface 730 and/or a pointing device such as a mouse as I/O interface 732. The keyboard and the mouse provide input to the computer 750.
The storage device 762 can consist of one or more of the following: a floppy disk, a hard disk drive, a magneto-optical disk drive, CD-ROM, magnetic tape or any other of a number of existing non-volatile storage devices. The flash disk or flash memory 610 is included as a storage device accessed through a respective device driver.
The software may be stored in a computer readable medium, including the storage device 762, or downloaded from a remote location via the interface 764 and communications channel 740 from the Internet 720 or another network location or site.
The computer system 700 includes the computer readable medium having such software or program code recorded such that instructions of the software or the program code can be carried out.
The computer system 700 is provided for illustrative purposes and other configurations can be used. The foregoing is merely an example of the types of computers or computer systems with which the described techniques and arrangements may be performed. The driver 620 is resident as a software component (such as an executable file, or dynamic link library). Intermediate storage of the program code of the device driver 620 and any data including entities, tickets, and the like may be accomplished using the memory 770, possibly in conjunction with the storage device 762.
In some instances, the device driver 620 may be supplied encoded on a CD-ROM or a floppy disk (both generally depicted by the storage device 762), or alternatively could be read by the user from the network via a modem device connected to the computer 750. Still further, the computer system 700 can load the device driver from other computer readable media. This may include magnetic tape, a ROM or integrated circuit, a magneto-optical disk, a radio or infra-red transmission channel between the computer and another device, a computer readable card such as a PCMCIA card, and the Internet 720 and Intranets including email transmissions and information recorded on Internet sites and the like. The foregoing are merely examples of relevant computer readable media. Other computer readable media may be used as appropriate.
The device driver 620, as a computer program, includes an expression, in any language, code or notation, of a set of instructions intended to cause a system having an information processing capability to perform a particular fiction on interacting with a respective associated device, either directly or after either or both of the following: a) conversion to another language, code or notation or b) reproduction in a different material form.
The described technique relates to an algorithm for emulating a disk in flash memory. The design aims to minimize erases and writes to the flash and thus speed up flash writes and increase flash life. This is a solution at the device driver level and is transparent to the file-system. This technique defines a policy to make use of a list of sector caches to buffer accesses to the flash-disk by recording certain characteristics of data access to the file-system blocks constituting the cached sectors, and dynamically calculating a weightage value. The technique also implements a procedure that automatically detects sectors that hold critical meta-data information or frequently accessed files, and gives preferential treatment to those detected sectors.
While various techniques and arrangements are described in the context of flash memory, the described techniques and arrangements relate more broadly to any memory medium having similar write, read, and erase characteristics as flash memories.